An MCP server that lets Claude Code, Codex, Cursor and other agents deploy your apps

Updated 2026-10-28

Agents Work For Me is cloud hosting built for AI coding agents. One MCP server gives your agent everything it needs to put a project online, while you keep control of the money.

Connect

No token to copy: the client signs in through the browser (OAuth) and you pick the project in the console.

Claude Code — then restart it and run /mcp → awfm → Authenticate

claude mcp add --transport http awfm https://mcp.agentsworkfor.me/mcp

Codex — the second command opens the browser

codex mcp add awfm --url https://mcp.agentsworkfor.me/mcp
codex mcp login awfm

Cursor and other clients (mcp.json)

{ "mcpServers": { "awfm": { "url": "https://mcp.agentsworkfor.me/mcp" } } }

What the agent can do

Servers — create_machine (SSH key, public IPv4, keepRunning), wait_operation, extend_lifetime, delete_machine and more.
Static sites, free — publish_static uploads HTML/CSS/JS and images and returns https://<name>.agentsworkforme.com in seconds, no server.
Public HTTPS address — publish_site puts a server’s port 80 at https://<name>.agentsworkforme.com.
Databases — create_database (PostgreSQL), get_database_credentials, rotate_database_password.
Money — estimate_cost, get_budget: the agent sees the price before creating anything.
Guides — get_skill publish-project returns the complete “put my project online” flow.

Built-in guardrails

Prepaid balance and per-project budget caps.
Every resource has a lifetime; “keep running” extends it only while money allows.
Tokens are bound to one project with explicit scopes and can be revoked any time.

FAQ

Which clients are supported?

Any MCP client with Streamable HTTP: Claude Code, Codex, Cursor, Windsurf, Cline and others. Clients with MCP OAuth sign in through the browser; others use a key from the console.

Can the agent overspend?

No. Every resource reserves its cost up front from a prepaid balance and is checked against the project budget; when money or limits run out the server refuses and the agent gets a clear error.

Is there a guide the agent can read?

Yes — /llms.txt describes the service and the full deploy flow for agents, and the get_skill tool returns step-by-step skills such as publish-project.